… Attaining efficiency and costs in targeted poverty alleviation programmes
Poverty targeting interventions in social development initiatives increased in popularity in the 1990s, due to the combination of evidence of high leakage in universal schemes together with political pressure to limit tax collection and reduce state expenditures that undermined governments’ ability to fund large universal programs. Nor is it a coincidence that the theory of selectivity in social provision grew stronger along with the rise of the neoliberal ideological shift in the 1980s and 1990s. The shift in development theory, supported by the World Bank and other international financial institutions such as the International Monetary Fund (IMF) or Inter- American Development Bank, was also reflected in development cooperation and aid programs. But it was not only external factors that led to the crisis of universalism and the shift to selective targeting programs; internal factors in developing countries, including economic mismanagement, bureaucratization, and corruption, also contributed to policy change in social development programs.
Targeting is often identified as more equitable and progressive than universal policies that transfer resources equally to all members of society. In theory, limited resources earmarked for social transfers would be used most efficiently when allocated to a designated subgroup of the population, generally excluding those not in need and concentrating resources on the poor and/or the extremely poor. There are several approaches to means-tested targeting such as targeting by category (for instance, age, disability, or employment status), targeting by means (e.g., assessment of income or assets), geographic targeting, or targeting by needs (e.g., particular circumstances).
In addition to not giving supplementary resources to those who already have sufficient means, the idea behind the use of targeting is to generate a pro-poor distribution of social services in society. Thus, the advantage is doubled: targeting would result in more poverty alleviation with the same or the lower total amount of social spending. Improved identification and targeting of the poor would then make social spending more effective without increasing the local tax burden (or the need for development aid). Identification of the poor could be done in various ways, from simple self-targeting models to, for instance, more complicated poverty mapping models or proxy means testing and even a combination of various models. But, as in many other fields, theory and practice can differ considerably.
Table 1: Efficacious targeting and Targeting errors
Covered Not Covered
Poor Successful targeting Exclusion error (under coverage)
Non-Poor Inclusion error (leakage) Successful exclusion
Even the strongest supporters and advocates of targeting recognise that its accuracy will never be perfect. Some of the poor will always be excluded from transfers, while some of the non-poor will become beneficiaries (see table 1). Due to information gaps and insufficient data to define the poor, precise targeting cannot be achieved. Misreporting may also lead to exclusion and inclusion errors. In addition to the problem of mistargeting, the process of identifying the poor requires extra resources, and such costs must be added to identification processes (for instance, the administration of targeting schemes, continuous updating of tools for the identification of the poor, fraud control and resource transfer costs). Using more advanced tools to identify the poor would minimize errors, but would also further increase costs. Finally, there are other but no less important, non-economic costs involved with targeting such as decreased political support for targeting programs when the population receiving benefits becomes smaller and less powerful, stigmatization of the poor, incentive gaps, corruption, and clientelism. Taking these factors into account is necessary for any serious evaluation of the effectiveness of targeting, and to move away from theory and toward evaluating practical effects. Another factor that contributes to misleading conclusions about the effectiveness of targeting schemes is how the data are presented. Evaluations of targeting schemes often focus on how much more the benefited poorest quintile or group defined as poor have gained from the “targeting” scheme than they would have with a neutral, universal scheme; or they draw attention to the large part of transferred benefits that go to the poorest quintile or quintiles.
Moreover, evaluation reports of targeted schemes generally draw attention to leakage, which are the funds that are mistakenly given to the non-poor. On that basis, it could be inferred that minimizing leakage should result in further reductions in social funding without cutting back on benefits for the poor, a win-win situation that conveniently fits into a broader vision of reduced state spending and fiscal restraint.
The much-cited report on targeting from the empirical works of Coady et al. in 2004 presented evidence in targeting efficiency and outcomes based on an evaluation of 122 antipoverty targeting interventions in 48 countries in various parts of the world using different targeting techniques. This empirical work has been used as a reference in the area of targeting, mainly because of the large number of studies on which it is based, and showed that the median targeting program transferred 25 percent more to poor individuals than a universal program. The 10 best performing schemes, of which the majority are in the Americas, demonstrate results whereby two to four times more resources were transferred to the poor than would have occurred under a universal scheme.
Nonetheless, there are serious weaknesses in these findings. To begin with, Coady et al. in their work in 2004 found that up to 25 percent of all targeting schemes have proven to be regressive, transferring fewer resources to the poor than under universal schemes. In these cases, what was initially designed to target the poor, in practice, has ended up targeting resources to the non-poor. The worst performers were food subsidy schemes; 6 of the 10 worst performers corresponded to this group. Furthermore, Coady et al. noted that there appears to be a negative correlation between low gross domestic product (GDP); often equal to low institutional capacity, and successful targeting schemes. The vast majority of the best performers in the study were located in the least poor countries and, consequently, have a more developed public administrative capacity.
Other underlying reasons explain why a targeted social program could be regressive, including weak and incomplete identification processes of the poor, caste, and class interests that influence the distribution of resources, the wrong geographical distribution of targeted services, and self-targeting type schemes that end up being attractive also to the non-poor. As noted above, one in four targeting schemes fails so badly that it ends up being regressive. If one focuses instead on targeting schemes that are identified as successful cases by targeting advocates, the results are still doubtful. The empirical evidence suggests that targeting fails to target the most vulnerable; for many of these programs, the excluded may constitute very large groups, and in some cases include the majority of the poor.
Part of the explanation lies in the fact that reporting targeting efficiency often includes only those poor receiving benefitsand not all of the poor. In this context, a statement such as the “median targeting program transfers 25 percent more to poor individuals than a universal program” reflects another reality and demonstrates just one of many innovative ways in which statistical data can be reported.
Table 2: Targeting Efficiency
Source: Author’s Computations
* 50 percent of the poor are excluded from benefits.
This paradox is explained further in Table 2, which compares social provision through four different simulations (of which two are ideal targeting situations, one relates to coverage and one relates to leakage). Assuming that in a fixed budget 1,000 units are allocated for social programs and given that there are four people in each quintile group, a universal scheme would allocate 50 units per person. In an ideally targeted scheme where 40 percent of the population qualifies as poor, the poor would thus receive 2.5 times more or 125 units. The argument for targeting often stops here.
However, if one goes beyond theory and take into account a “standard” ratio of a successful targeting scheme, the image becomes more complex. Because of incomplete information, in practice, there would always be some leakage and under coverage. For instance, persons who should qualify as poor are sometimes not identified as such and as a result do not receive benefits. The fourth example in the table assumes a program with 32 percent leakage and 50 percent under coverage, a fairly average number for targeted social programs in developing countries, as illustrated in the following examples.
The program has a targeting efficiency of 3.4 times more resources to the poor than an average scheme or, as presented in another (common) way of reporting, two-thirds of resources go to the poorest 40 percent, which appears to give most benefits to the poor. However, a large part of this “efficiency” is a direct result of the exclusion of the poor from social programs; half of the eligible poor receive no benefits at all. The third example in table 2 shows that the targeting efficiency rate would decrease dramatically if all of the poor were included in the scheme, reaching levels that are closer to the results of the universal scheme.
The levels of under-coverage and leakage mentioned above are not extreme cases and could have been taken from a random evaluation of any targeted social program. A World Bank report described several “successful” social programs in the Americas where targeting efficiency in the individual studies varies between 1.68 and 4.0 in relation to universal schemes, demonstrating that social programs transfer considerably more resources to the poor than universal schemes. The programs are depicted as examples that support targeting and confirm earlier studies on targeting efficiency by the World Bank, for which the numbers cited above are presented as first-hand evidence.